Research On Power Transmission Line Fault Diagnosis Based On Convolutional Neural Network

For a country’s economic development,homeland security and people’s happiness,electricity security is very important.Even if the power system fails for a short time,the loss to national defense security,social production and people’s life is hard to measure by money.The fault of power transmission line is one of the important reasons for the fault of power system.Therefore,after the fault of power transmission line,it is meaningful to study the fault diagnosis.The traditional fault diagnosis methods of power transmission line mainly rely on experience and theoretical derivation.Although it has strong explanability,with the increasing complexity of power system and the increasing diversification of power equipment,the traditional fault diagnosis methods are faced with problems such as the difficulty of expression of theoretical formula and poor stability.In the face of complex fault problems,the traditional fault diagnosis methods lack of diagnosis experience,and the traditional fault diagnosis methods are difficult to solve the problem of multiple positions of power transmission line simultaneous fault.Depth in order to solve the above problems,this dissertation,by using the neural network in feature extraction and advantages of nonlinear expression,automatic extraction of power transmission line fault characteristics,with a fault diagnosis model and complete the power transmission line fault feature extraction and fault diagnosis,reduce the separately from the traditional method of fault feature extraction and fault diagnosis for the diagnosis of error.In this dissertation,the three-phase voltage,three-phase current and their zero-sequence components in the power system are taken as the characteristic signals.Wavelet packet transform is used to analyze the characteristics of the signals in the time-frequency domain,and compared with the time-domain characteristics and Fourier characteristics.Based on the theory of deep learning,this dissertation designs the Convolutional Neural Network(CNN)fault diagnosis model.By contrast,Stacked Autoencoders(SAE)and Deep Belief Network(DBN)fault diagnosis models are designed simultaneously.The structural composition and basic working principle of the three fault diagnosis models are analyzed.Aiming at the learning rate setting problem of the traditional DBN,an improved DBN with adaptive learning rate is adopted.Aiming at the problems of fixed nodes and weak generalization performance of traditional CNN,the dropout mechanism was introduced and the improved CNN was adopted as the fault diagnosis model of power transmission line.In this dissertation,the IEEE 14-node and the IEEE 30-node standard grid structure are respectively taken as examples to build a power transmission line fault simulation model.Multiple fault points and sampling points are set up,and the electrical volume of sampling points is output as the characteristic signal data set.By adding artificial Gaussian random noise to the characteristic signal data set,the power transmission line fault sample set is obtained by secondary processing of the data set.Through multi-angle comparative experiments,the influence of model parameters such as learning rate,dropout value and batch sample number on diagnosis accuracy was studied,and the reasonable setting interval of model parameters was determined.The diagnosis effect of the fault diagnosis model is studied when the fault occurs simultaneously in multiple positions of power transmission line.The change of diagnostic accuracy with the SNR of fault sample set is studied.The influence of power system topological structure change on fault diagnosis accuracy is studied.The processing method of non-equilibrium data set is studied.The differences of diagnostic accuracy of three fault diagnosis models were compared.Is adopted in this dissertation,the experimental results show that the power transmission line fault diagnosis model based on CNN has high diagnostic accuracy.Both power line transmission short circuit fault and power transmission line open circuit fault are considered,a failure occurs at the same time in the multi position has better diagnosis capability,and has good noise resistance,have wider application prospect.